FluxyChat

Features

Empathy layer (voice)

Opt-in prosody detection adapts agent tone and escalation without telling the user they seem stressed.

Empathy layer (voice)

During voice sessions, optional prosody analysis (pitch, rate, pauses) on the client feeds a coarse emotion signal to the agent runtime. The agent adjusts tone and escalation hints. It never announces that the user sounds frustrated.

Console

Voice AI (/voice-ai): enable empathy per room in the settings panel.

Settings: room-level empathy toggles stored in room_empathy_settings.

Opt-in

  • Per-room toggle in D1.
  • Raw prosody is not persisted. Only an aggregated signal with TTL in KV.
  • Low confidence means no prompt change.

Client

useEmpathyProsody hook sends:

POST /rooms/:roomId/empathy/signal

Body: coarse bucket (calm, tense, elevated) plus confidence, not raw audio.

Agent runtime

When a signal is present, agent-runtime appends a short empathy block to the system prompt (adjust tone, offer pause) without exposing detection to the user.

Voice stages

Discord Stages-style speaker and listener roles with active-speaker detection in voice rooms.

Multi-agent debate

Two or three perspective agents debate visibly before a moderator synthesizes the final answer.

On this page

Empathy layer (voice)ConsoleOpt-inClientAgent runtimeRelated